Claude Mythos Preview
Per Anthropic, an AI model that is very good at finding security weaknesses in software. Because it could be misused, it was not released to the public, only to selected organizations.
Claude Mythos Preview is a general-purpose AI model made by Anthropic. Anthropic says it is especially strong at computer security work: it can find weaknesses in software and even build ways to use them in a real attack. Think of a highly skilled inspector who tests the lock on every door. In a burglar's hands the same skill would be dangerous, so Anthropic says it does not plan to release the model generally and has opened it first only to organizations taking part in Project Glasswing.
Watch out Examples such as the 27-year-old OpenBSD flaw, and the counts of findings, are Anthropic's own statements. Most of the flaws are not yet fixed, so details are withheld. What a small business can do is not memorize the model's name but keep its website and tools updated without delay.
Source: Claude Mythos Preview's cybersecurity capabilities · Project Glasswing: Securing critical software for the AI era (checked 2026-10-26)